Implementation of 3D FFTs across multiple GPUs in shared memory environments

Nimalan Nandapalan, Jiri Jaros, Alistair P. Rendell, Bradley Treeby

Research output: Chapter in Book/Report/Conference proceedingConference contribution

6 Citations (Scopus)

Abstract

In this paper, a novel implementation of the distributed 3D Fast Fourier Transform (FFT) on a multi-GPU platform using CUDA is presented. The 3D FFT is the core of many simulation methods, thus its fast calculation is critical. The main bottleneck of the distributed 3D FFT is the global data exchange which must be performed. The latest version of CUDA introduces direct GPU-to-GPU transfers using a Unified Virtual Address space (UVA) that provides new possibilities for optimising the communication part of the FFT. Here, we propose different implementations of the distributed 3D FFT, investigate their behaviour, and compare their performance with the single GPU CUFFT and CPU-based FFTW libraries. In particular, we demonstrate the advantage of direct GPU-to-GPU transfers over data exchanges via host main memory. Our preliminary results show that running the distributed 3D FFT with four GPUs can bring a 12% speedup over the single node (CUFFT) while also enabling the calculation of 3D FFTs of larger datasets. Replacing the global data exchange via shared memory with direct GPU-to-GPU transfers reduces the execution time by up to 49%. This clearly shows that direct GPU-to-GPU transfers are the key factor in obtaining good performance on multi-GPU systems.

Original languageEnglish
Title of host publicationProceedings - 13th International Conference on Parallel and Distributed Computing, Applications, and Technologies, PDCAT 2012
Pages167-172
Number of pages6
DOIs
Publication statusPublished - 14 Dec 2012
Externally publishedYes
Event13th International Conference on Parallel and Distributed Computing, Applications, and Technologies, PDCAT 2012 - Beijing, China
Duration: 14 Dec 201216 Dec 2012

Publication series

NameParallel and Distributed Computing, Applications and Technologies, PDCAT Proceedings

Conference

Conference13th International Conference on Parallel and Distributed Computing, Applications, and Technologies, PDCAT 2012
CountryChina
CityBeijing
Period14/12/1216/12/12

Keywords

  • distributed
  • FFT
  • multi-GPU
  • shared-memory
  • unified-virtual-address
  • UVA

Fingerprint Dive into the research topics of 'Implementation of 3D FFTs across multiple GPUs in shared memory environments'. Together they form a unique fingerprint.

  • Cite this

    Nandapalan, N., Jaros, J., Rendell, A. P., & Treeby, B. (2012). Implementation of 3D FFTs across multiple GPUs in shared memory environments. In Proceedings - 13th International Conference on Parallel and Distributed Computing, Applications, and Technologies, PDCAT 2012 (pp. 167-172). [6589258] (Parallel and Distributed Computing, Applications and Technologies, PDCAT Proceedings). https://doi.org/10.1109/PDCAT.2012.79